438 



Fishery Bulletin 92(2). 1994 



Data type 



"product variable," to see if it con- 

 tributed additional information for 

 determining dolphin habitats. In 

 summary, the six oceanographic 

 variables included were 1) surface 

 temperature, TEMP; 2) surface 

 salinity, SAL; 3) surface density, 

 SIGMAT; 4) thermocline depth, 

 Z20; 5) thermocline strength, ZD; 

 and 6) chlorophyll, LOGC. 



We examined the effects of 

 interannual variability by includ- 

 ing years (scaled 1-5) as categori- 

 cal variables (details below). Ad- 

 ditionally, we examined the contri- 

 bution of fixed geographic effects 

 by including latitude and longi- 

 tude in some analyses. All environmental variables 

 (oceanographic and geographic) were normalized 

 prior to multivariate analyses to remove effects from 

 differing scales of measurement. 



Relationships between dolphin school distributions 

 and environmental variation were analyzed by us- 

 ing canonical correspondence analysis (CCA; Ter 

 Braak, 1986). We used the computer program 

 CANOCO (Ter Braak, 1985). Correspondence analy- 

 sis is an eigenvector ordination technique, similar to 

 principal components analysis, that can be used to 

 investigate community structure. These methods 

 extract dominant, orthogonal axes of variation in 



Table 2 



Oceanographic data from the Monitoring of Porpoise Stocks expedition, 

 1986-90, used in the canonical correspondence analyses. Table entries list 

 numbers of observations for discrete measurements, or number of km cov- 

 ered during continuous measurements. XBT = expendable bathythermo- 

 graph; CTD = conductivity-temperature-depth. 



1986 



1987 



1988 



1989 



1990 



Total 



Surface temperature, 

 salinity (km) 



Surface chlorophyll 

 measurements 



XBT measurements 

 (drops) 



CTD measurements 

 (stations) 



28,917 27,735 24,224 27,323 32,398 140,597 



3.763 



1,144 



244 



1,927 



1,160 



280 



3.613 3,552 4,448 



835 



352 



778 



352 



809 



368 



7,303 



4,726 



1,596 



abundance indices for multiple species at multiple 

 sites. Typically, the ordination axes are then inter- 

 preted indirectly with the help of external knowledge 

 and data on environmental gradients, either quali- 

 tatively or with regression methods (Gauch, 1982). 

 In contrast to principal components analysis and 

 other linear methods, correspondence analysis (CA, 

 also called reciprocal averaging) fits nonlinear 

 Gaussian (unimodal) models to the species abun- 

 dance data. Canonical correspondence analysis is an 

 extension of CA in which the species ordination is 

 done directly and iteratively in relation to environ- 

 mental variables. CCA is an efficient ordination tech- 



